https://bugzilla.kernel.org/show_bug.cgi?id=207959 --- Comment #3 from Luc Van Oostenryck (luc.vanoostenryck@xxxxxxxxx) --- (In reply to Asher Gordon from comment #2) > Perhaps '-Wno-universal-initializer' should be the default? Well, that's really the point. The problem Sparse also gives the warnings corresponding to clang's -Wnonnull and my understanding is that these warnings are desired for the kernel even when coming from using '{ 0 }'. > > My very personal point of view is that the correct syntax should be '{ }' > > because it conveys much better the idea of a default initializer. This > > single zero in '{ 0 }' is just confusing. > > I can see your point, but unfortunately, as Ramsay Jones says here[1] and > Alexander Monakov here[2], this is not standard C. So '{ }' isn't an option > if we want to be portable. Yes, I know, it's a pity. It's why I said 'should be'. > Andrew Pinski's suggestion[3] is also an option, > but that seems ugly to me. Yes, it's far from ideal. > I'm writing a library, Mu[4], which has a structure for which the > 'designated_init' attribute is appropriate (see the 'MU_OPT' structure > here[5]). However, I don't want to force my users not to use '{ 0 }', which > is why I think this feature would be useful. Interesting. Yes, I understand. Git was in the same kind of situation, it's why I added '-Wno-universal-initializer'. Can't you just add this option in your SPARSE_FLAGS or something like that? > Also, a minor note: In the test program I attached, the attribute needs to > be specified after the closing brace to work with Sparse. Yes, it's a known problem. Sparse accept 'type attributes' (those situated just after the keyword 'struct', 'union' or 'enum') but ignore them. I've some unfinished patches for this ... since some time already :( -- You are receiving this mail because: You are watching the assignee of the bug.
